Status : Verified
| Personal Name | Boquio, Eujene Nikka V. |
|---|---|
| Resource Title | Beyond Traditional Automated Essay Scoring: A Deep Learning Approach with the Unified BERT-based Adaptive Trait Essay Scoring System |
| Date Issued | 04 July 2024 |
| Abstract | This thesis addresses the challenging yet relevant task of Automated Essay Scoring (AES) that automates the evaluation of written student work. Traditional AES systems often struggle with adaptability to new or unseen prompts and typically require extensive manual feature engineering. Furthermore, cross-prompt trait scoring, a relatively underexplored area in AES, presents additional challenges in evaluating essays across diverse prompts while maintaining accuracy in trait-specific feedback. To overcome these limitations, this research introduces the Unified BERT-based Adaptive Trait Essay Scoring System (UBATESS), a novel end-to-end deep learning model that leverages the capabilities of the sophisticated BERT architecture. UBATESS incorporates advanced machine learning techniques including hybrid pooling strategies to utilize both single- and multi-layer outputs of BERT, domain adaptation through pre-training on and joint learning with auxiliary tasks, and multi-task learning to simultaneously score multiple essay traits. Furthermore, the model integrates prompt information to improve prompt-specific scoring performance and employs novel text processing techniques to effectively manage long essays. Experimental results show that UBATESS significantly outperforms existing models in cross-prompt trait scoring, with an improvement of 10.9% on the average agreement QWK metric over the previous state-of-the-art model, establishing UBATESS as a state-of-the-art solution in the field of AES. UBATESS is a unified model that demonstrates a superior ability to generalize across different prompts and provides multi-dimensional feedback with holistic and trait-specific scores, making it highly effective for diverse educational applications. |
| Degree Course | MS Computer Science |
| Language | English |
| Keyword | artificial intelligence, nlp, deep learning, essay grading |
| Material Type | Thesis/Dissertation |
Preliminary Pages
634.41 Kb
Category : F - Regular work, i.e., it has no patentable invention or creation, the author does not wish for personal publication, there is no confidential information.
Access Permission : Open Access